Few-Shot Event Detection Based on Multi-scale Feature Fusion and Multi-way Prototype Network

This research offers a few-shot event detection technique based on multi-scale feature fusion and multi-way prototype network to address the issue of inadequate prototype representation learning caused by limited training data and insufficient feature information extraction. First, the standard convolutional network is improved by the multi-scale attention module, and each sentence's feature information is improved by the local and global feature fusion approach. Second, to address the imbalance of class and difficulty samples, a multi-way prototype network is shown. In the 5-way-5-shot context, the suggested model enhances the F1 value in the FewEvent dataset by 3.38% when compared to the robust baseline model of few-shot event detection. Experiments show that the robustness and stability of the proposed model are significantly improved, and the performance is also significantly improved.